Prediction of miRNA-disease association based on heterogeneous hypergraph convolution and heterogeneous graph multi-scale convolution

• Published in: Health information science and systems 2025-12, Vol.13 (1), p.4, Article 4
• Main Authors: Dai, Wei, Pang, Sifan, He, Zhichen, Fu, Xiaodong, Liu, Li, Liu, Lijun, Yu, Ning
• Format: Article
• Language: English
• Subjects: Bioinformatics; Computational Biology/Bioinformatics; Computer Science; Convolution; Disease; Graph theory; Graphical representations; Graphs; Health Informatics; Information Systems and Communication Service; MicroRNAs; Regularization
• ISSN: 2047-2501; 2047-2501
• DOI: 10.1007/s13755-024-00319-1

Making the accurate prediction of miRNA-disease associations essential for medical interventions. Current computational models often fail to capture the complexity of miRNA-disease associations. This study proposes HHMDA, a method based on heterogeneous hypergraph convolution and heterogeneous graph multi-scale convolution, to predict the association between miRNA and disease. Firstly, HHMDA constructs a heterogeneous graph of miRNA-disease relationships. Then, a graph convolution is run on the heterogeneous graph to capture the multi-scale feature representations of miRNA and disease. MiRNA-disease association are reconstructed based on these features. Meanwhile, HHMDA constructs a heterogeneous hypergraph with miRNAs and diseases as nodes, and the hyperedges consist of miRNAs and diseases linked to the same genes. HHMDA performs hypergraph graph convolution operation on the heterogeneous hypergraph to extract the high-order features of miRNA and disease. Finally, these features are leveraged to calculate the Laplacian regularization loss and combined with the miRNA-disease association matrix reconstruction loss to optimize the model. The experimental results show that HHMDA has advantages over the existing state-of-the-art methods under different experimental settings.